Chapter 18 Exercises: Fundamentals Models

Conceptual Exercises

Exercise 18.1 — Economic Variables

For each of the following economic indicators, explain (a) what it measures, (b) why it might predict election outcomes, and (c) one limitation of using it as a forecasting variable:

1. Real disposable personal income growth per capita
2. Unemployment rate (level)
3. Change in unemployment rate
4. Consumer price index (inflation)
5. Real GDP growth in the second quarter of the election year
6. Consumer confidence index

Which of these do you think is the strongest predictor? Why? What combination might be most powerful?

Exercise 18.2 — The Time for Change Model

The Time for Change model uses three variables: Q2 GDP growth, presidential net approval in June, and a first-term incumbency dummy.

Using the data below (hypothetical), apply the TFC model structure to predict the incumbent party's popular vote share. Assume the model equation is approximately:

Incumbent party vote share = 48.0 + (0.54 × Q2 GDP growth) + (0.10 × net approval) + (2.5 × first-term incumbent dummy)

a) Fill in the predicted vote shares. b) In which scenarios does the incumbent party win? Lose? (Assume they win if predicted share > 50%) c) How sensitive are the predictions to the GDP growth variable versus the approval variable? d) What's the effect of being a first-term incumbent in these scenarios?

Exercise 18.3 — Retrospective Voting Logic

V.O. Key described voters as "rational gods of vengeance and reward." Apply this logic:

a) Identify three economic or policy conditions that voters might "reward" an incumbent for. b) Identify three conditions they might "punish" an incumbent for. c) How does the retrospective voting framework explain why a president can have good economic numbers but still low approval ratings? d) What are the limits of retrospective voting as a theory of democratic accountability?

Exercise 18.4 — Incumbency Advantage

The incumbency advantage in House elections has declined from roughly 5-8 points in the 1970s-80s to roughly 2-4 points today.

a) Using the three explanations offered in the chapter (partisan sorting, nationalization, media fragmentation), explain in your own words how each one would reduce the incumbency advantage. b) Are there any factors that might be increasing incumbency advantages in some contexts today? c) How would you measure the incumbency advantage empirically? What data would you need? d) If the incumbency advantage reaches zero, what does that imply about how congressional elections work?

Analytical Exercises

Exercise 18.5 — Model Comparison

The chapter describes several different fundamentals models (TFC, Bread and Peace, Lewis-Beck, Holbrook). For each model:

a) What is the primary theoretical argument underlying the model? b) What variables does it use? c) What are its key advantages relative to the others? d) What are its key limitations?

Then answer: If you were building a fundamentals model for a Senate race (not a presidential race), which elements would you retain, and what would you add?

Exercise 18.6 — The 2022 Midterms and Structural Models

In 2022, structural models (based on Biden's low approval ratings and high inflation) predicted significant Republican gains. Republicans did gain seats but not as many as predicted, and they underperformed in several key Senate races.

Research the 2022 midterms and answer: a) What were the major structural indicators heading into the election? b) What did major fundamentals-style models predict? c) What actually happened? d) What factors might explain the gap between structural predictions and actual outcomes? e) Does the 2022 case undermine fundamentals models, or is it within their expected error range?

Exercise 18.7 — State-Level Structural Modeling

You want to build a structural model for a specific state's Senate race. List all the variables you would consider, organized by category: - National economic variables - State-level economic variables - Presidential approval (national) - Presidential/party performance in the state (historical) - Incumbent senator's approval (state-level) - Demographic trends in the state

For each variable, specify: (a) where you would get the data, (b) what time period you'd use, and (c) what theoretical reason connects it to the Senate race outcome.

Applied Exercises

Exercise 18.8 — Historical Fundamentals Test

Select three presidential elections from 1976-2016. For each: a) Gather the Q2 GDP growth rate, June presidential approval, and incumbency status b) Apply the TFC model equation from Exercise 18.2 c) Compare the prediction to the actual popular vote result d) For any election where the model was significantly off, identify potential explanations

What is the average error across your three elections? How does this compare to the uncertainty in individual polls?

Exercise 18.9 — Generic Ballot Analysis

Using publicly available data (RCP, 538 archives), gather generic ballot readings for the last three election cycles (2018, 2020, 2022) at four points in time: one year before the election, six months before, three months before, and the final average.

a) How did the generic ballot evolve in each cycle? b) Did early generic ballot readings predict the direction of election outcomes? c) What was the final generic ballot, and how did it compare to the actual national House popular vote? d) What limitations of the generic ballot as a forecasting tool do your findings illustrate?

Discussion Questions

Discussion 18.1

If fundamentals models can predict presidential elections with reasonable accuracy using only economic data available in the summer of the election year, what does this imply about the role of campaign advertising, debates, and media coverage? Does this finding make you more or less interested in the mechanics of political campaigns?

Discussion 18.2

The "Time for Change" model was developed by estimating a regression on data from post-WWII elections. As of today, that's somewhere between 18-20 elections — a small sample. How confident should we be in a model estimated on 18 observations? What are the risks of overfitting? How would you test whether the model is capturing a genuine structural regularity versus a historical coincidence?

Discussion 18.3

The chapter describes how the 2020 COVID-19 pandemic complicated fundamentals models that use Q2 GDP growth. When a structural shock of this kind occurs — one that no forecasting model could anticipate and one where voters may not attribute economic conditions to the incumbent — what should a good forecaster do? Should they abandon the model, adjust it, or continue to use it with explicit acknowledgment of the anomaly?

Discussion 18.4

Consider the prediction vs. explanation distinction. A fundamentals model predicts that the incumbent party will win 51.8% of the vote. The actual result is 52.1%. The model was "right." But does the model explain why the incumbent won? What would an explanation require that a prediction does not?